<b>Monthly Brix measurements and optimization of near-infrared spectral preprocessing for Brix prediction in sugar beet using regression analysis</b>

<p dir="ltr"><b>1. Monthly Brix measurement of sugar beet juice</b></p><p dir="ltr">Monthly Brix values of sugar beet juice were recorded from July to October. Juice was extracted from cube-shaped pieces cut slightly below the crown area of the sugar beet, specifically in the proximal and distal groove regions. No observable differences in Brix values were observed between the groove-distal and groove-proximal regions (Table S1).</p><p><br></p><p dir="ltr">Table S1 Monthly Brix of sugar beet juice</p><p dir="ltr">(EAEF180306_Table_S1.csv)</p><p dir="ltr"><i>n</i>: number of samples, StDev: standard deviation.</p><p dir="ltr"><b>2. Selection of optimal Savitzky–Golay filter settings for Brix prediction in sugar beet using LinkSquare near-infrared (NIR) spectra and partial least squares (PLS) regression analysis</b></p><p dir="ltr">Table S2 shows the optimal Savitzky–Golay filter settings for Brix prediction in sugar beets based on spectra acquired with the LinkSquare NIR spectrometer and analyzed using PLS regression. Increasing the polynomial order did not improve prediction accuracy for either intact or peeled spectra. For intact spectra, a 0-order polynomial fit combined with a 21-point filter yielded the best performance. In contrast, for peeled spectra, the optimal result was achieved using a 0-order polynomial fit with a 17-point filter length.</p><p><br></p><p dir="ltr">Table S2 Comparison of PLS regression results across Savitzky–Golay filter settings using LinkSquare NIR spectra for Brix prediction in sugar beet</p><p dir="ltr">(EAEF180306_Table_S2.csv)</p><p dir="ltr">LV: latent variable, <i>R</i>²: coefficient of determination, RMSEC: root mean square error of calibration, RMSECV: root mean square error of cross-validation, RMSEP: root mean square error of prediction, RPD: ratio of performance to deviation, IWT: individual-without treatment, IS: individual-smoothing.</p><p dir="ltr"><b>3. Selection of optimal Savitzky–Golay filter settings for Brix prediction in sugar beet using LinkSquare NIR spectra and multiple linear regression (MLR) analysis</b></p><p dir="ltr">Table S3 presents the optimal Savitzky–Golay filter settings for Brix prediction in sugar beet based on spectra acquired with the LinkSquare NIR spectrometer and analyzed using MLR. Consistent with the PLS regression results, increasing the polynomial order did not improve prediction accuracy for either intact or peeled spectra. For intact spectra, a 0-order polynomial fit combined with a 21-point filter yielded the best performance. For peeled spectra, the optimal result was achieved using a 0-order polynomial fit and a 17-point filter.</p><p dir="ltr"><br></p><p dir="ltr">Table S3 Comparison of MLR results across Savitzky–Golay filter settings using LinkSquare NIR spectra for Brix prediction in sugar beet</p><p dir="ltr">(EAEF180306_Table_S3.csv)</p><p dir="ltr"><b>4. Selection of optimal Savitzky–Golay filter windows for Brix prediction in sugar beet using second-derivative spectra from S-G1 NIR spectrometer and PLS regression</b></p><p dir="ltr">Table S4 presents the optimal Savitzky–Golay filter windows for Brix prediction in sugar beet based on second-derivative spectra measured using the S-G1 NIR spectrometer and analyzed using PLS regression. Because second-derivative preprocessing requires a fixed polynomial order of two, only the filter window was varied. A 2nd-order polynomial fit combined with a 21-point filter window produced optimal results for both intact and peeled spectra.</p><p><br></p><p dir="ltr">Table S4 Comparison of PLS regression results across Savitzky–Golay filter windows using second-derivative spectra from the S-G1 NIR spectrometer for Brix prediction in sugar beet</p><p dir="ltr">(EAEF180306_Table_S4.csv)</p><p dir="ltr">ID2: individual with second derivative.</p><p dir="ltr"><b>5. Selection of optimal Savitzky–Golay filter windows for Brix prediction in sugar beet using second derivative spectra from S-G1 NIR spectrometer and MLR</b></p><p dir="ltr">Table S5 presents the optimal Savitzky–Golay filter windows for Brix prediction in sugar beets based on second-derivative spectra measured using the S-G1 NIR spectrometer and analyzed using MLR. Similar to the PLS regression results, a 2nd-order polynomial fit combined with a 21-point filter window produced optimal results for both intact and peeled spectra.</p><p dir="ltr"><br></p><p dir="ltr">Table S5 Comparison of MLR results across Savitzky–Golay filter windows using second-derivative spectra from the S-G1 NIR spectrometer for Brix prediction in sugar beet</p><p dir="ltr">(EAEF180306_Table_S5.csv)</p><p dir="ltr"><b>6. Selection of optimal pretreatment methods for Brix prediction in sugar beet using LinkSquare NIR spectra and PLS regression</b></p><p dir="ltr">Table S6 summarizes the results of common spectral pretreatment methods. The findings indicated that Standard Normal Variate (SNV) and Multiplicative Scatter Correction were the most effective methods for both intact and peeled sugar beet spectra measured using the LinkSquare NIR spectrometer.</p><p dir="ltr"><br></p><p dir="ltr">Table S6 Comparison of PLS regression results across pretreatment methods using LinkSquare NIR spectra for Brix prediction in sugar beet</p><p dir="ltr">(EAEF180306_Table_S6.csv)</p><p dir="ltr">MSC: multiplicative scatter correction, ISSNV: individual-smoothing with SNV, ISMSC: individual-smoothing with MSC, ISD1: individual-smoothing with first derivative, ISD2: individual-smoothing with second derivative.</p><p dir="ltr"><b>7. Selection of optimal pretreatment methods for Brix prediction in sugar beet using LinkSquare NIR spectra and MLR</b></p><p dir="ltr">Table S7 shows the results of common spectral pretreatment methods. The findings indicated that SNV was the most effective method for both intact and peeled sugar beet spectra measured using a LinkSquare NIR spectrometer.</p><p dir="ltr"><br></p><p dir="ltr">Table S7 Comparison of MLR results across pretreatment methods using LinkSquare NIR spectra for Brix prediction in sugar beet</p><p dir="ltr">(EAEF180306_Table_S7.csv)</p><p dir="ltr"><b>8. Selection of optimal pretreatment methods for Brix prediction in sugar beet using S-G1 NIR spectra and PLS regression</b></p><p dir="ltr">Table S8 summarizes the results of common spectral pretreatment methods. The findings indicated that the second-derivative method was the most effective for both intact and peeled sugar beet spectra measured with the S-G1 NIR spectrometer.</p><p dir="ltr"><br></p><p dir="ltr">Table S8 Comparison of PLS regression results across pretreatment methods using S-G1 NIR spectra for Brix prediction in sugar beet</p><p dir="ltr">(EAEF180306_Table_S8.csv)</p><p dir="ltr"><b>9. Selection of optimal pretreatment methods for Brix prediction in sugar beet using S-G1 NIR spectra and MLR</b></p><p dir="ltr">Table S9 shows the results of common spectral preprocessing methods. Similar to the PLS regression results, the findings indicated that the second-derivative method was the most effective for both intact and peeled sugar beet spectra measured using the S-G1 NIR spectrometer.</p><p dir="ltr"><br></p><p dir="ltr">Table S9 Comparison of MLR results across pretreatment methods using S-G1 NIR spectra for Brix prediction in sugar beet</p><p dir="ltr">(EAEF180306_Table_S9.csv)</p>